Yang Lu

Predicting the elastic properties of multiphase concrete remains challenging due to its inherent heterogeneity and nonlinear behavior. This study proposes a community-enhanced graph neural network (CE-GNN) to predict the elastic response of four-phase cementitious materials. Three-dimensional microstructures obtained from X-ray CT images are transformed into graphs using a Mapper clustering algorithm, and community detection is incorporated into the GNN architecture to capture localized mechanical interactions. Ground-truth elastic properties are generated through finite element-based homogenization. Results show that CE-GNN achieves an R 2 of 0.97 for bulk modulus prediction and reduces the end-to-end wall-clock runtime by approximately 57%–68% compared with FEM homogenization on the same workstation, while outperforming both CNN and standard GNN baselines.

Pavement surface defects such as potholes pose significant safety risks and accelerate infrastructure deterioration. Accurate and automated detection of such defects requires both advanced sensing technologies and robust deep learning models. In this study, we propose PoFormer, a Transformer–CNN hybrid framework designed for precise segmentation of pavement potholes from heterogeneous image datasets. The architecture leverages the global feature extraction ability of Transformers and the fine-grained localization capability of CNNs, achieving superior segmentation accuracy compared to state-of-the-art models. To construct a representative dataset, we combined open source images with high-resolution field data acquired using a multi-sensor pavement inspection vehicle equipped with a line-scan camera and infrared/laser-assisted lighting. This sensing system provides millimeter-level resolution and continuous 3D surface imaging under diverse environmental conditions, ensuring robust training inputs for deep learning. Experimental results demonstrate that PoFormer achieves a mean IoU of 77.23% and a mean pixel accuracy of 84.48%, outperforming existing CNN-based models. By integrating multi-sensor data acquisition with advanced hybrid neural networks, this work highlights the potential of 3D imaging and sensing technologies for intelligent pavement condition monitoring and automated infrastructure maintenance.

This study presents a comprehensive framework for evaluating community-level flood resilience by integrating the fragility of individual buildings, the functionality of critical infrastructure sectors, and their interdependencies. Using performance-based engineering principles, the framework quantifies resilience through isolated building fragility curves, sector-specific functionality fragility curves, and a synthesized community-level functionality model. Applied to a virtual community of 1000 archetypal buildings, the analysis reveals that community functionality decreases with increasing flood depth, reaching a critical threshold of 0.87 at 1.57 m. The sensitivity analysis underscores the importance of accounting for intersectoral dependencies, as they significantly influence community-wide functionality. The results highlight the residential sector’s dominant role in shaping resilience and its cascading effects on other sectors. This framework provides actionable insights for planners and stakeholders, emphasizing the need to prioritize interventions in sectors with the highest vulnerability and dependency to enhance disaster preparedness and response strategies. This framework, novel in its integration of building-level fragility curves with community-wide intersectoral dependencies, provides actionable insights for planners and stakeholders, emphasizing targeted interventions in vulnerable sectors to enhance flood resilience.

The interface between old and new concrete is a critical component in many construction practices, including concrete pavements, bridge decks, hydraulic dams, and buildings undergoing rehabilitation. Despite various treatments to enhance bonding, this interface often remains a weak layer that compromises overall structural performance. Traditional design methods typically oversimplify the interface as a homogeneous or empirically adjusted factor, resulting in significant uncertainties. This paper introduces a novel framework for quantifying the anisotropic properties of old–new concrete interfaces using X-ray computed tomography (CT) and finite element-based numerical homogenization. The elastic coefficient matrix reveals that specimens away from the interface exhibit higher values in both normal and shear directions, with normal direction values averaging 33.15% higher and shear direction values 39.96% higher than those at the interface. A total of 10 sampling units along the interface were collected and analyzed to identify the “weakest vectors” in normal and shear directions. The “weakest vectors” at the interface show consistent orientations with an average cosine similarity of 0.62, compared with an average cosine similarity of 0.23 at the non-interface, which demonstrates directional features. Conversely, the result of average cosine similarity at the interface shows randomness that originates from the anisotropy of materials. The average angle between normal and shear stresses was found to be 88.64°, indicating a predominantly orthogonal relationship, though local stress distributions introduced slight deviations. These findings highlight the importance of understanding the anisotropic properties of old–new concrete interfaces to improve design and rehabilitation practices in concrete and structural engineering.

The accurate prediction of pavement performance is essential for transportation administration or management to appropriately allocate resources road maintenance and upkeep. The international roughness index (IRI) is one of the most commonly used pavement performance indicators to reflect the surface roughness. However, the existing research on IRI prediction mainly focuses on using linear regression or traditional machine learning, which cannot take into account the historical effects of IRI caused by climate, traffic, pavement construction and intermittent maintenance. In this work, a long short-term memory (LSTM)-based model, LSTM+MA, is proposed to predict the IRI of pavements using the time-series data extracted from the long-term pavement performance (LTPP) dataset. Effective preprocessing methods and hyperparameter fine-tuning are selected to improve the accuracy of the model. The performance of the LSTM+MA is compared with other state-of-the-art models, including logistic regressor (LR), support vector regressor (SVR), random forest (RF), K-nearest-neighbor regressor (KNR), fully connected neural network (FNN), XGBoost (XGB), recurrent neural network (RNN) and LSTM. The results show that selected preprocessing methods can help the model learn quickly from the data and reach high accuracy in small epochs. Also, it shows that the proposed LSTM+MA model significantly outperforms other models, with an R² of 0.965 and a mean square error (MSE) of 0.030 in the test datasets. Moreover, an overfitting score is proposed in this work to represent the severity degree of the overfitting problem, and it shows that the proposed model does not suffer severely from overfitting.

Research on wildfires and social vulnerability has gained significant importance due to the increasing frequency and severity of wildfires around the world. This study investigates the dynamic changes in social vulnerability to wildfires over a decade in Idaho, USA, utilizing GIS-based tools and a quasi-experimental design. We assess the evolving nature of social vulnerability at a local scale, emphasizing both spatial and temporal dynamics. Initially, we identified social vulnerability trends in relation to varying levels of wildfire risk. The research then employs propensity score matching to contrast areas affected by wildfires in 2012 with similar non-affected regions, thereby quantifying the short-term shifts in social vulnerability post-wildfires. The results indicate that regions with a high wildfire risk may display elevated vulnerability, characterized by an increase in unemployment rates and a reduction in high-income households. These findings tentatively demonstrate the compounded effect of wildfires on already vulnerable populations, highlighting the critical need for targeted interventions. Ultimately, this study underscores the importance of integrating dynamic social vulnerability assessments into wildfire management and planning, aiming to enhance community resilience and equitable resource distribution in the face of escalating wildfire threats.

Cement hydration kinetics, characterized by heat generation in early-age concrete, poses a modeling challenge. This work proposes a physics-informed neural network (PINN) named PINN-CHK designed for cement hydration kinetics, to predict early-age temperature rises in cement paste. PINN-CHK leverages data-driven solutions to craft a high-fidelity prediction model, encompassing material properties and maturity functions in cement hydration. Trained on heated cement paste data, it simultaneously fits experimental results and underlying physics, yielding a mesh-free simulation. Incorporating governing partial differential equations (PDEs), and initial and boundary conditions into its loss function, PINN-CHK architecture undergoes rigorous benchmark testing, demonstrating unparalleled predictive accuracy compared to conventional deep-learning methods. It excels in predicting complete temperature fields during spatial–temporal cement hydration, achieving a remarkable relative L2 error as low as 0.00341. PINN-CHK achieves exceptional convergence and accuracy with only 5% of the training data, ushering in a new era in this crucial field. This innovative approach bridges the gap between theory and practice, offering an attractive alternative to conventional finite element solvers for enhanced comprehension of cement hydration kinetics and concrete maturity and strength development in cement-based materials.